Spaces:
Sleeping
Sleeping
Create app.py
Browse files
app.py
ADDED
|
@@ -0,0 +1,171 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import os
|
| 2 |
+
import time
|
| 3 |
+
import random
|
| 4 |
+
from typing import Dict
|
| 5 |
+
|
| 6 |
+
from fastapi import FastAPI, Request, HTTPException
|
| 7 |
+
from fastapi.responses import StreamingResponse, HTMLResponse, JSONResponse
|
| 8 |
+
|
| 9 |
+
from ollama import Client
|
| 10 |
+
|
| 11 |
+
app = FastAPI()
|
| 12 |
+
|
| 13 |
+
# ββ CONFIG βββββββββββββββββββββββββββββββββββββββββββββ
|
| 14 |
+
MASTER_API_KEY = os.getenv("MASTER_API_KEY", "changeme")
|
| 15 |
+
BASE_URL = os.getenv("BASE_URL", "https://ollama.com")
|
| 16 |
+
|
| 17 |
+
# ββ LOAD KEYS ββββββββββββββββββββββββββββββββββββββββββ
|
| 18 |
+
def load_keys(prefix="OLLAMA_KEY_"):
|
| 19 |
+
pairs = []
|
| 20 |
+
for k, v in os.environ.items():
|
| 21 |
+
if k.startswith(prefix) and v:
|
| 22 |
+
try:
|
| 23 |
+
idx = int(k.replace(prefix, ""))
|
| 24 |
+
except:
|
| 25 |
+
idx = 9999
|
| 26 |
+
pairs.append((idx, v.strip()))
|
| 27 |
+
pairs.sort(key=lambda x: x[0])
|
| 28 |
+
return [v for _, v in pairs]
|
| 29 |
+
|
| 30 |
+
API_KEYS = load_keys()
|
| 31 |
+
print(f"[INIT] Loaded {len(API_KEYS)} keys")
|
| 32 |
+
|
| 33 |
+
# ββ STATE ββββββββββββββββββββββββββββββββββββββββββββββ
|
| 34 |
+
key_status: Dict[str, Dict] = {
|
| 35 |
+
k: {
|
| 36 |
+
"status": "active",
|
| 37 |
+
"fail": 0,
|
| 38 |
+
"cooldown": 0,
|
| 39 |
+
"req": 0,
|
| 40 |
+
"err": 0,
|
| 41 |
+
"last": "-"
|
| 42 |
+
}
|
| 43 |
+
for k in API_KEYS
|
| 44 |
+
}
|
| 45 |
+
|
| 46 |
+
# ββ AUTH βββββββββββββββββββββββββββββββββββββββββββββββ
|
| 47 |
+
def auth(req: Request):
|
| 48 |
+
if req.headers.get("Authorization") != f"Bearer {MASTER_API_KEY}":
|
| 49 |
+
raise HTTPException(401, "Unauthorized")
|
| 50 |
+
|
| 51 |
+
# ββ KEY PICKER βββββββββββββββββββββββββββββββββββββββββ
|
| 52 |
+
def pick_key():
|
| 53 |
+
now = time.time()
|
| 54 |
+
valid = []
|
| 55 |
+
|
| 56 |
+
for k, v in key_status.items():
|
| 57 |
+
if v["status"] == "dead":
|
| 58 |
+
continue
|
| 59 |
+
if v["cooldown"] > now:
|
| 60 |
+
continue
|
| 61 |
+
valid.append(k)
|
| 62 |
+
|
| 63 |
+
return random.choice(valid) if valid else None
|
| 64 |
+
|
| 65 |
+
# ββ STATUS UPDATE ββββββββββββββββββββββββββββββββββββββ
|
| 66 |
+
def mark_limit(k):
|
| 67 |
+
key_status[k]["status"] = "limited"
|
| 68 |
+
key_status[k]["cooldown"] = time.time() + 60
|
| 69 |
+
key_status[k]["err"] += 1
|
| 70 |
+
key_status[k]["last"] = "rate_limit"
|
| 71 |
+
|
| 72 |
+
def mark_dead(k, reason="dead"):
|
| 73 |
+
key_status[k]["status"] = "dead"
|
| 74 |
+
key_status[k]["err"] += 1
|
| 75 |
+
key_status[k]["last"] = reason
|
| 76 |
+
|
| 77 |
+
def mark_ok(k):
|
| 78 |
+
key_status[k]["status"] = "active"
|
| 79 |
+
key_status[k]["fail"] = 0
|
| 80 |
+
key_status[k]["last"] = "-"
|
| 81 |
+
|
| 82 |
+
# ββ STREAM PROXY βββββββββββββββββββββββββββββββββββββββ
|
| 83 |
+
@app.post("/chat")
|
| 84 |
+
async def chat(req: Request):
|
| 85 |
+
auth(req)
|
| 86 |
+
body = await req.json()
|
| 87 |
+
|
| 88 |
+
model = body.get("model")
|
| 89 |
+
messages = body.get("messages")
|
| 90 |
+
|
| 91 |
+
if not model:
|
| 92 |
+
return JSONResponse({"error": "model required"}, status_code=400)
|
| 93 |
+
|
| 94 |
+
if not messages:
|
| 95 |
+
return JSONResponse({"error": "messages required"}, status_code=400)
|
| 96 |
+
|
| 97 |
+
for _ in range(len(API_KEYS)):
|
| 98 |
+
key = pick_key()
|
| 99 |
+
|
| 100 |
+
if not key:
|
| 101 |
+
return JSONResponse({"error": "all keys exhausted"}, status_code=503)
|
| 102 |
+
|
| 103 |
+
try:
|
| 104 |
+
client = Client(
|
| 105 |
+
host=BASE_URL,
|
| 106 |
+
headers={"Authorization": f"Bearer {key}"}
|
| 107 |
+
)
|
| 108 |
+
|
| 109 |
+
key_status[key]["req"] += 1
|
| 110 |
+
|
| 111 |
+
def stream():
|
| 112 |
+
try:
|
| 113 |
+
for part in client.chat(model, messages=messages, stream=True):
|
| 114 |
+
yield part["message"]["content"]
|
| 115 |
+
mark_ok(key)
|
| 116 |
+
|
| 117 |
+
except Exception as e:
|
| 118 |
+
key_status[key]["fail"] += 1
|
| 119 |
+
key_status[key]["err"] += 1
|
| 120 |
+
key_status[key]["last"] = str(e)
|
| 121 |
+
|
| 122 |
+
if "429" in str(e):
|
| 123 |
+
mark_limit(key)
|
| 124 |
+
elif key_status[key]["fail"] >= 3:
|
| 125 |
+
mark_dead(key, "fail")
|
| 126 |
+
|
| 127 |
+
raise e
|
| 128 |
+
|
| 129 |
+
return StreamingResponse(stream(), media_type="text/plain")
|
| 130 |
+
|
| 131 |
+
except Exception:
|
| 132 |
+
continue
|
| 133 |
+
|
| 134 |
+
return JSONResponse({"error": "all keys failed"}, status_code=500)
|
| 135 |
+
|
| 136 |
+
# ββ DASHBOARD ββββββββββββββββββββββββββββββββββββββββββ
|
| 137 |
+
@app.get("/dashboard")
|
| 138 |
+
async def dash(req: Request):
|
| 139 |
+
auth(req)
|
| 140 |
+
|
| 141 |
+
now = time.time()
|
| 142 |
+
|
| 143 |
+
html = """
|
| 144 |
+
<html><body style="background:#0f172a;color:white;font-family:sans-serif">
|
| 145 |
+
<h2>KEY STATUS</h2>
|
| 146 |
+
<table border="1" cellpadding="8">
|
| 147 |
+
<tr>
|
| 148 |
+
<th>Key</th><th>Status</th><th>Req</th><th>Err</th><th>Cooldown</th><th>Last</th>
|
| 149 |
+
</tr>
|
| 150 |
+
"""
|
| 151 |
+
|
| 152 |
+
for k, v in key_status.items():
|
| 153 |
+
cd = max(0, int(v["cooldown"] - now))
|
| 154 |
+
html += f"""
|
| 155 |
+
<tr>
|
| 156 |
+
<td>{k[:6]}...</td>
|
| 157 |
+
<td>{v['status']}</td>
|
| 158 |
+
<td>{v['req']}</td>
|
| 159 |
+
<td>{v['err']}</td>
|
| 160 |
+
<td>{cd}s</td>
|
| 161 |
+
<td>{v['last']}</td>
|
| 162 |
+
</tr>
|
| 163 |
+
"""
|
| 164 |
+
|
| 165 |
+
html += "</table></body></html>"
|
| 166 |
+
return HTMLResponse(html)
|
| 167 |
+
|
| 168 |
+
# ββ ROOT βββββββββββββββββββββββββββββββββββββββββββββββ
|
| 169 |
+
@app.get("/")
|
| 170 |
+
def root():
|
| 171 |
+
return {"status": "ok", "keys": len(API_KEYS)}
|